Both approaches submerge servers in a dielectric fluid. They diverge on whether that fluid is meant to boil, and almost every practical difference follows from that one decision.
Single-phase keeps the fluid liquid
Heat is carried away by circulating warm fluid to a heat exchanger. The fluid stays in one state, which keeps tank sealing, vapor management, and fluid loss comparatively simple, and it is where the supplier landscape in our records is deepest.
Two-phase uses the boiling point as the control
A fluid engineered to boil at a low temperature removes heat as it changes state, which can hold chip temperatures in a narrow band at high flux. The cost is vapor containment, condenser design, and a tighter dependence on fluid supply and regulatory treatment.
Supplier depth is not symmetric
Across the companies tracked here, single-phase immersion is offered by specialists including GRC, Submer, Asperitas, and Midas, while two-phase activity in the market has concentrated on direct-to-chip rather than full immersion. Availability, not physics, is often the deciding factor for a first deployment.
Decide on the service model
Ask how a failed server is removed, how fluid is drained, filtered, tested, and topped up, what happens to fluid at end of life, and who is contractually responsible for each step. An immersion deployment lives or dies on those procedures.
- Hardware warranty position for submerged components
- Fluid analysis cadence and replacement cost
- Vapour or evaporation losses at operating temperature
- Local disposal and reporting obligations
Fluids and the total cost of the cooling system
In a single-phase immersion cooling system the dielectric fluid is typically a hydrocarbon or synthetic oil that stays in liquid form and is comparatively inexpensive to replace. Two-phase cooling depends on engineered fluids whose cost, availability, and regulatory treatment have all moved in recent years. Because the fluid is a consumable rather than a one-time purchase, a data center comparing the two architectures should model fluid volume, expected annual losses, filtration, and replacement over the life of the hall rather than at commissioning only.
Density and hardware compatibility
Both approaches support high-density racks well beyond what air can carry, and neither is limited by airflow. The practical constraint is hardware: server components are qualified for air or for a specific liquid, and submerging a system that was not designed for it puts the warranty, and often the reliability, in question. Confirm which server models the supplier has validated, whether hard drives and optical transceivers are excluded, and what the server vendor will support in writing.